Wow imagine the odds that you could train an Opus 4.6 class model in just 3 months with only $25 million and do such a good job that your model itself thinks it’s Claude Opus in the reasoning trace. There’s no other explanation for that except just being better at AI than those lazy Americans.
Quantian
Open weight gemini 5/opus 4.6 trained solely on Huawei Ascend chips. Its so over
